1 // 2016 and later: Unicode, Inc. and others. 2 // License & terms of use: http://www.unicode.org/copyright.html 3 /* 4 ******************************************************************************* 5 * 6 * Copyright (C) 2009-2014, International Business Machines 7 * Corporation and others. All Rights Reserved. 8 * 9 ******************************************************************************* 10 * file name: normalizer2impl.h 11 * encoding: UTF-8 12 * tab size: 8 (not used) 13 * indentation:4 14 * 15 * created on: 2009nov22 16 * created by: Markus W. Scherer 17 */ 18 19 #ifndef __NORMALIZER2IMPL_H__ 20 #define __NORMALIZER2IMPL_H__ 21 22 #include "unicode/utypes.h" 23 24 #if !UCONFIG_NO_NORMALIZATION 25 26 #include "unicode/normalizer2.h" 27 #include "unicode/unistr.h" 28 #include "unicode/unorm.h" 29 #include "unicode/utf16.h" 30 #include "mutex.h" 31 #include "uset_imp.h" 32 #include "utrie2.h" 33 34 U_NAMESPACE_BEGIN 35 36 struct CanonIterData; 37 38 class ByteSink; 39 class Edits; 40 class InitCanonIterData; 41 class LcccContext; 42 43 class U_COMMON_API Hangul { 44 public: 45 /* Korean Hangul and Jamo constants */ 46 enum { 47 JAMO_L_BASE=0x1100, /* "lead" jamo */ 48 JAMO_L_END=0x1112, 49 JAMO_V_BASE=0x1161, /* "vowel" jamo */ 50 JAMO_V_END=0x1175, 51 JAMO_T_BASE=0x11a7, /* "trail" jamo */ 52 JAMO_T_END=0x11c2, 53 54 HANGUL_BASE=0xac00, 55 HANGUL_END=0xd7a3, 56 57 JAMO_L_COUNT=19, 58 JAMO_V_COUNT=21, 59 JAMO_T_COUNT=28, 60 61 JAMO_VT_COUNT=JAMO_V_COUNT*JAMO_T_COUNT, 62 63 HANGUL_COUNT=JAMO_L_COUNT*JAMO_V_COUNT*JAMO_T_COUNT, 64 HANGUL_LIMIT=HANGUL_BASE+HANGUL_COUNT 65 }; 66 67 static inline UBool isHangul(UChar32 c) { 68 return HANGUL_BASE<=c && c<HANGUL_LIMIT; 69 } 70 static inline UBool 71 isHangulLV(UChar32 c) { 72 c-=HANGUL_BASE; 73 return 0<=c && c<HANGUL_COUNT && c%JAMO_T_COUNT==0; 74 } 75 static inline UBool isJamoL(UChar32 c) { 76 return (uint32_t)(c-JAMO_L_BASE)<JAMO_L_COUNT; 77 } 78 static inline UBool isJamoV(UChar32 c) { 79 return (uint32_t)(c-JAMO_V_BASE)<JAMO_V_COUNT; 80 } 81 static inline UBool isJamoT(UChar32 c) { 82 int32_t t=c-JAMO_T_BASE; 83 return 0<t && t<JAMO_T_COUNT; // not JAMO_T_BASE itself 84 } 85 static UBool isJamo(UChar32 c) { 86 return JAMO_L_BASE<=c && c<=JAMO_T_END && 87 (c<=JAMO_L_END || (JAMO_V_BASE<=c && c<=JAMO_V_END) || JAMO_T_BASE<c); 88 } 89 90 /** 91 * Decomposes c, which must be a Hangul syllable, into buffer 92 * and returns the length of the decomposition (2 or 3). 93 */ 94 static inline int32_t decompose(UChar32 c, UChar buffer[3]) { 95 c-=HANGUL_BASE; 96 UChar32 c2=c%JAMO_T_COUNT; 97 c/=JAMO_T_COUNT; 98 buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT); 99 buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT); 100 if(c2==0) { 101 return 2; 102 } else { 103 buffer[2]=(UChar)(JAMO_T_BASE+c2); 104 return 3; 105 } 106 } 107 108 /** 109 * Decomposes c, which must be a Hangul syllable, into buffer. 110 * This is the raw, not recursive, decomposition. Its length is always 2. 111 */ 112 static inline void getRawDecomposition(UChar32 c, UChar buffer[2]) { 113 UChar32 orig=c; 114 c-=HANGUL_BASE; 115 UChar32 c2=c%JAMO_T_COUNT; 116 if(c2==0) { 117 c/=JAMO_T_COUNT; 118 buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT); 119 buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT); 120 } else { 121 buffer[0]=orig-c2; // LV syllable 122 buffer[1]=(UChar)(JAMO_T_BASE+c2); 123 } 124 } 125 private: 126 Hangul(); // no instantiation 127 }; 128 129 class Normalizer2Impl; 130 131 class U_COMMON_API ReorderingBuffer : public UMemory { 132 public: 133 /** Constructs only; init() should be called. */ 134 ReorderingBuffer(const Normalizer2Impl &ni, UnicodeString &dest) : 135 impl(ni), str(dest), 136 start(NULL), reorderStart(NULL), limit(NULL), 137 remainingCapacity(0), lastCC(0) {} 138 /** Constructs, removes the string contents, and initializes for a small initial capacity. */ 139 ReorderingBuffer(const Normalizer2Impl &ni, UnicodeString &dest, UErrorCode &errorCode); 140 ~ReorderingBuffer() { 141 if(start!=NULL) { 142 str.releaseBuffer((int32_t)(limit-start)); 143 } 144 } 145 UBool init(int32_t destCapacity, UErrorCode &errorCode); 146 147 UBool isEmpty() const { return start==limit; } 148 int32_t length() const { return (int32_t)(limit-start); } 149 UChar *getStart() { return start; } 150 UChar *getLimit() { return limit; } 151 uint8_t getLastCC() const { return lastCC; } 152 153 UBool equals(const UChar *start, const UChar *limit) const; 154 UBool equals(const uint8_t *otherStart, const uint8_t *otherLimit) const; 155 156 UBool append(UChar32 c, uint8_t cc, UErrorCode &errorCode) { 157 return (c<=0xffff) ? 158 appendBMP((UChar)c, cc, errorCode) : 159 appendSupplementary(c, cc, errorCode); 160 } 161 // s must be in NFD, otherwise change the implementation. 162 UBool append(const UChar *s, int32_t length, 163 uint8_t leadCC, uint8_t trailCC, 164 UErrorCode &errorCode); 165 UBool appendBMP(UChar c, uint8_t cc, UErrorCode &errorCode) { 166 if(remainingCapacity==0 && !resize(1, errorCode)) { 167 return FALSE; 168 } 169 if(lastCC<=cc || cc==0) { 170 *limit++=c; 171 lastCC=cc; 172 if(cc<=1) { 173 reorderStart=limit; 174 } 175 } else { 176 insert(c, cc); 177 } 178 --remainingCapacity; 179 return TRUE; 180 } 181 UBool appendZeroCC(UChar32 c, UErrorCode &errorCode); 182 UBool appendZeroCC(const UChar *s, const UChar *sLimit, UErrorCode &errorCode); 183 void remove(); 184 void removeSuffix(int32_t suffixLength); 185 void setReorderingLimit(UChar *newLimit) { 186 remainingCapacity+=(int32_t)(limit-newLimit); 187 reorderStart=limit=newLimit; 188 lastCC=0; 189 } 190 void copyReorderableSuffixTo(UnicodeString &s) const { 191 s.setTo(ConstChar16Ptr(reorderStart), (int32_t)(limit-reorderStart)); 192 } 193 private: 194 /* 195 * TODO: Revisit whether it makes sense to track reorderStart. 196 * It is set to after the last known character with cc<=1, 197 * which stops previousCC() before it reads that character and looks up its cc. 198 * previousCC() is normally only called from insert(). 199 * In other words, reorderStart speeds up the insertion of a combining mark 200 * into a multi-combining mark sequence where it does not belong at the end. 201 * This might not be worth the trouble. 202 * On the other hand, it's not a huge amount of trouble. 203 * 204 * We probably need it for UNORM_SIMPLE_APPEND. 205 */ 206 207 UBool appendSupplementary(UChar32 c, uint8_t cc, UErrorCode &errorCode); 208 void insert(UChar32 c, uint8_t cc); 209 static void writeCodePoint(UChar *p, UChar32 c) { 210 if(c<=0xffff) { 211 *p=(UChar)c; 212 } else { 213 p[0]=U16_LEAD(c); 214 p[1]=U16_TRAIL(c); 215 } 216 } 217 UBool resize(int32_t appendLength, UErrorCode &errorCode); 218 219 const Normalizer2Impl &impl; 220 UnicodeString &str; 221 UChar *start, *reorderStart, *limit; 222 int32_t remainingCapacity; 223 uint8_t lastCC; 224 225 // private backward iterator 226 void setIterator() { codePointStart=limit; } 227 void skipPrevious(); // Requires start<codePointStart. 228 uint8_t previousCC(); // Returns 0 if there is no previous character. 229 230 UChar *codePointStart, *codePointLimit; 231 }; 232 233 /** 234 * Low-level implementation of the Unicode Normalization Algorithm. 235 * For the data structure and details see the documentation at the end of 236 * this normalizer2impl.h and in the design doc at 237 * http://site.icu-project.org/design/normalization/custom 238 */ 239 class U_COMMON_API Normalizer2Impl : public UObject { 240 public: 241 Normalizer2Impl() : normTrie(NULL), fCanonIterData(NULL) { 242 fCanonIterDataInitOnce.reset(); 243 } 244 virtual ~Normalizer2Impl(); 245 246 void init(const int32_t *inIndexes, const UTrie2 *inTrie, 247 const uint16_t *inExtraData, const uint8_t *inSmallFCD); 248 249 void addLcccChars(UnicodeSet &set) const; 250 void addPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const; 251 void addCanonIterPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const; 252 253 // low-level properties ------------------------------------------------ *** 254 255 UBool ensureCanonIterData(UErrorCode &errorCode) const; 256 257 uint16_t getNorm16(UChar32 c) const { return UTRIE2_GET16(normTrie, c); } 258 259 UNormalizationCheckResult getCompQuickCheck(uint16_t norm16) const { 260 if(norm16<minNoNo || MIN_YES_YES_WITH_CC<=norm16) { 261 return UNORM_YES; 262 } else if(minMaybeYes<=norm16) { 263 return UNORM_MAYBE; 264 } else { 265 return UNORM_NO; 266 } 267 } 268 UBool isAlgorithmicNoNo(uint16_t norm16) const { return limitNoNo<=norm16 && norm16<minMaybeYes; } 269 UBool isCompNo(uint16_t norm16) const { return minNoNo<=norm16 && norm16<minMaybeYes; } 270 UBool isDecompYes(uint16_t norm16) const { return norm16<minYesNo || minMaybeYes<=norm16; } 271 272 uint8_t getCC(uint16_t norm16) const { 273 if(norm16>=MIN_NORMAL_MAYBE_YES) { 274 return getCCFromNormalYesOrMaybe(norm16); 275 } 276 if(norm16<minNoNo || limitNoNo<=norm16) { 277 return 0; 278 } 279 return getCCFromNoNo(norm16); 280 } 281 static uint8_t getCCFromNormalYesOrMaybe(uint16_t norm16) { 282 return (uint8_t)(norm16 >> OFFSET_SHIFT); 283 } 284 static uint8_t getCCFromYesOrMaybe(uint16_t norm16) { 285 return norm16>=MIN_NORMAL_MAYBE_YES ? getCCFromNormalYesOrMaybe(norm16) : 0; 286 } 287 uint8_t getCCFromYesOrMaybeCP(UChar32 c) const { 288 if (c < minCompNoMaybeCP) { return 0; } 289 return getCCFromYesOrMaybe(getNorm16(c)); 290 } 291 292 /** 293 * Returns the FCD data for code point c. 294 * @param c A Unicode code point. 295 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 296 */ 297 uint16_t getFCD16(UChar32 c) const { 298 if(c<minDecompNoCP) { 299 return 0; 300 } else if(c<=0xffff) { 301 if(!singleLeadMightHaveNonZeroFCD16(c)) { return 0; } 302 } 303 return getFCD16FromNormData(c); 304 } 305 /** 306 * Returns the FCD data for the next code point (post-increment). 307 * Might skip only a lead surrogate rather than the whole surrogate pair if none of 308 * the supplementary code points associated with the lead surrogate have non-zero FCD data. 309 * @param s A valid pointer into a string. Requires s!=limit. 310 * @param limit The end of the string, or NULL. 311 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 312 */ 313 uint16_t nextFCD16(const UChar *&s, const UChar *limit) const { 314 UChar32 c=*s++; 315 if(c<minDecompNoCP || !singleLeadMightHaveNonZeroFCD16(c)) { 316 return 0; 317 } 318 UChar c2; 319 if(U16_IS_LEAD(c) && s!=limit && U16_IS_TRAIL(c2=*s)) { 320 c=U16_GET_SUPPLEMENTARY(c, c2); 321 ++s; 322 } 323 return getFCD16FromNormData(c); 324 } 325 /** 326 * Returns the FCD data for the previous code point (pre-decrement). 327 * @param start The start of the string. 328 * @param s A valid pointer into a string. Requires start<s. 329 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 330 */ 331 uint16_t previousFCD16(const UChar *start, const UChar *&s) const { 332 UChar32 c=*--s; 333 if(c<minDecompNoCP) { 334 return 0; 335 } 336 if(!U16_IS_TRAIL(c)) { 337 if(!singleLeadMightHaveNonZeroFCD16(c)) { 338 return 0; 339 } 340 } else { 341 UChar c2; 342 if(start<s && U16_IS_LEAD(c2=*(s-1))) { 343 c=U16_GET_SUPPLEMENTARY(c2, c); 344 --s; 345 } 346 } 347 return getFCD16FromNormData(c); 348 } 349 350 /** Returns TRUE if the single-or-lead code unit c might have non-zero FCD data. */ 351 UBool singleLeadMightHaveNonZeroFCD16(UChar32 lead) const { 352 // 0<=lead<=0xffff 353 uint8_t bits=smallFCD[lead>>8]; 354 if(bits==0) { return false; } 355 return (UBool)((bits>>((lead>>5)&7))&1); 356 } 357 /** Returns the FCD value from the regular normalization data. */ 358 uint16_t getFCD16FromNormData(UChar32 c) const; 359 360 /** 361 * Gets the decomposition for one code point. 362 * @param c code point 363 * @param buffer out-only buffer for algorithmic decompositions 364 * @param length out-only, takes the length of the decomposition, if any 365 * @return pointer to the decomposition, or NULL if none 366 */ 367 const UChar *getDecomposition(UChar32 c, UChar buffer[4], int32_t &length) const; 368 369 /** 370 * Gets the raw decomposition for one code point. 371 * @param c code point 372 * @param buffer out-only buffer for algorithmic decompositions 373 * @param length out-only, takes the length of the decomposition, if any 374 * @return pointer to the decomposition, or NULL if none 375 */ 376 const UChar *getRawDecomposition(UChar32 c, UChar buffer[30], int32_t &length) const; 377 378 UChar32 composePair(UChar32 a, UChar32 b) const; 379 380 UBool isCanonSegmentStarter(UChar32 c) const; 381 UBool getCanonStartSet(UChar32 c, UnicodeSet &set) const; 382 383 enum { 384 // Fixed norm16 values. 385 MIN_YES_YES_WITH_CC=0xfe02, 386 JAMO_VT=0xfe00, 387 MIN_NORMAL_MAYBE_YES=0xfc00, 388 JAMO_L=2, // offset=1 hasCompBoundaryAfter=FALSE 389 INERT=1, // offset=0 hasCompBoundaryAfter=TRUE 390 391 // norm16 bit 0 is comp-boundary-after. 392 HAS_COMP_BOUNDARY_AFTER=1, 393 OFFSET_SHIFT=1, 394 395 // For algorithmic one-way mappings, norm16 bits 2..1 indicate the 396 // tccc (0, 1, >1) for quick FCC boundary-after tests. 397 DELTA_TCCC_0=0, 398 DELTA_TCCC_1=2, 399 DELTA_TCCC_GT_1=4, 400 DELTA_TCCC_MASK=6, 401 DELTA_SHIFT=3, 402 403 MAX_DELTA=0x40 404 }; 405 406 enum { 407 // Byte offsets from the start of the data, after the generic header. 408 IX_NORM_TRIE_OFFSET, 409 IX_EXTRA_DATA_OFFSET, 410 IX_SMALL_FCD_OFFSET, 411 IX_RESERVED3_OFFSET, 412 IX_RESERVED4_OFFSET, 413 IX_RESERVED5_OFFSET, 414 IX_RESERVED6_OFFSET, 415 IX_TOTAL_SIZE, 416 417 // Code point thresholds for quick check codes. 418 IX_MIN_DECOMP_NO_CP, 419 IX_MIN_COMP_NO_MAYBE_CP, 420 421 // Norm16 value thresholds for quick check combinations and types of extra data. 422 423 /** Mappings & compositions in [minYesNo..minYesNoMappingsOnly[. */ 424 IX_MIN_YES_NO, 425 /** Mappings are comp-normalized. */ 426 IX_MIN_NO_NO, 427 IX_LIMIT_NO_NO, 428 IX_MIN_MAYBE_YES, 429 430 /** Mappings only in [minYesNoMappingsOnly..minNoNo[. */ 431 IX_MIN_YES_NO_MAPPINGS_ONLY, 432 /** Mappings are not comp-normalized but have a comp boundary before. */ 433 IX_MIN_NO_NO_COMP_BOUNDARY_BEFORE, 434 /** Mappings do not have a comp boundary before. */ 435 IX_MIN_NO_NO_COMP_NO_MAYBE_CC, 436 /** Mappings to the empty string. */ 437 IX_MIN_NO_NO_EMPTY, 438 439 IX_MIN_LCCC_CP, 440 IX_RESERVED19, 441 IX_COUNT 442 }; 443 444 enum { 445 MAPPING_HAS_CCC_LCCC_WORD=0x80, 446 MAPPING_HAS_RAW_MAPPING=0x40, 447 // unused bit 0x20, 448 MAPPING_LENGTH_MASK=0x1f 449 }; 450 451 enum { 452 COMP_1_LAST_TUPLE=0x8000, 453 COMP_1_TRIPLE=1, 454 COMP_1_TRAIL_LIMIT=0x3400, 455 COMP_1_TRAIL_MASK=0x7ffe, 456 COMP_1_TRAIL_SHIFT=9, // 10-1 for the "triple" bit 457 COMP_2_TRAIL_SHIFT=6, 458 COMP_2_TRAIL_MASK=0xffc0 459 }; 460 461 // higher-level functionality ------------------------------------------ *** 462 463 // NFD without an NFD Normalizer2 instance. 464 UnicodeString &decompose(const UnicodeString &src, UnicodeString &dest, 465 UErrorCode &errorCode) const; 466 /** 467 * Decomposes [src, limit[ and writes the result to dest. 468 * limit can be NULL if src is NUL-terminated. 469 * destLengthEstimate is the initial dest buffer capacity and can be -1. 470 */ 471 void decompose(const UChar *src, const UChar *limit, 472 UnicodeString &dest, int32_t destLengthEstimate, 473 UErrorCode &errorCode) const; 474 475 const UChar *decompose(const UChar *src, const UChar *limit, 476 ReorderingBuffer *buffer, UErrorCode &errorCode) const; 477 void decomposeAndAppend(const UChar *src, const UChar *limit, 478 UBool doDecompose, 479 UnicodeString &safeMiddle, 480 ReorderingBuffer &buffer, 481 UErrorCode &errorCode) const; 482 UBool compose(const UChar *src, const UChar *limit, 483 UBool onlyContiguous, 484 UBool doCompose, 485 ReorderingBuffer &buffer, 486 UErrorCode &errorCode) const; 487 const UChar *composeQuickCheck(const UChar *src, const UChar *limit, 488 UBool onlyContiguous, 489 UNormalizationCheckResult *pQCResult) const; 490 void composeAndAppend(const UChar *src, const UChar *limit, 491 UBool doCompose, 492 UBool onlyContiguous, 493 UnicodeString &safeMiddle, 494 ReorderingBuffer &buffer, 495 UErrorCode &errorCode) const; 496 497 /** sink==nullptr: isNormalized() */ 498 UBool composeUTF8(uint32_t options, UBool onlyContiguous, 499 const uint8_t *src, const uint8_t *limit, 500 ByteSink *sink, icu::Edits *edits, UErrorCode &errorCode) const; 501 502 const UChar *makeFCD(const UChar *src, const UChar *limit, 503 ReorderingBuffer *buffer, UErrorCode &errorCode) const; 504 void makeFCDAndAppend(const UChar *src, const UChar *limit, 505 UBool doMakeFCD, 506 UnicodeString &safeMiddle, 507 ReorderingBuffer &buffer, 508 UErrorCode &errorCode) const; 509 510 UBool hasDecompBoundaryBefore(UChar32 c) const; 511 UBool norm16HasDecompBoundaryBefore(uint16_t norm16) const; 512 UBool hasDecompBoundaryAfter(UChar32 c) const; 513 UBool norm16HasDecompBoundaryAfter(uint16_t norm16) const; 514 UBool isDecompInert(UChar32 c) const { return isDecompYesAndZeroCC(getNorm16(c)); } 515 516 UBool hasCompBoundaryBefore(UChar32 c) const { 517 return c<minCompNoMaybeCP || norm16HasCompBoundaryBefore(getNorm16(c)); 518 } 519 UBool hasCompBoundaryAfter(UChar32 c, UBool onlyContiguous) const { 520 return norm16HasCompBoundaryAfter(getNorm16(c), onlyContiguous); 521 } 522 UBool isCompInert(UChar32 c, UBool onlyContiguous) const { 523 uint16_t norm16=getNorm16(c); 524 return isCompYesAndZeroCC(norm16) && 525 (norm16 & HAS_COMP_BOUNDARY_AFTER) != 0 && 526 (!onlyContiguous || isInert(norm16) || *getMapping(norm16) <= 0x1ff); 527 } 528 529 UBool hasFCDBoundaryBefore(UChar32 c) const { return hasDecompBoundaryBefore(c); } 530 UBool hasFCDBoundaryAfter(UChar32 c) const { return hasDecompBoundaryAfter(c); } 531 UBool isFCDInert(UChar32 c) const { return getFCD16(c)<=1; } 532 private: 533 friend class InitCanonIterData; 534 friend class LcccContext; 535 536 UBool isMaybe(uint16_t norm16) const { return minMaybeYes<=norm16 && norm16<=JAMO_VT; } 537 UBool isMaybeOrNonZeroCC(uint16_t norm16) const { return norm16>=minMaybeYes; } 538 static UBool isInert(uint16_t norm16) { return norm16==INERT; } 539 static UBool isJamoL(uint16_t norm16) { return norm16==JAMO_L; } 540 static UBool isJamoVT(uint16_t norm16) { return norm16==JAMO_VT; } 541 uint16_t hangulLVT() const { return minYesNoMappingsOnly|HAS_COMP_BOUNDARY_AFTER; } 542 UBool isHangulLV(uint16_t norm16) const { return norm16==minYesNo; } 543 UBool isHangulLVT(uint16_t norm16) const { 544 return norm16==hangulLVT(); 545 } 546 UBool isCompYesAndZeroCC(uint16_t norm16) const { return norm16<minNoNo; } 547 // UBool isCompYes(uint16_t norm16) const { 548 // return norm16>=MIN_YES_YES_WITH_CC || norm16<minNoNo; 549 // } 550 // UBool isCompYesOrMaybe(uint16_t norm16) const { 551 // return norm16<minNoNo || minMaybeYes<=norm16; 552 // } 553 // UBool hasZeroCCFromDecompYes(uint16_t norm16) const { 554 // return norm16<=MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT; 555 // } 556 UBool isDecompYesAndZeroCC(uint16_t norm16) const { 557 return norm16<minYesNo || 558 norm16==JAMO_VT || 559 (minMaybeYes<=norm16 && norm16<=MIN_NORMAL_MAYBE_YES); 560 } 561 /** 562 * A little faster and simpler than isDecompYesAndZeroCC() but does not include 563 * the MaybeYes which combine-forward and have ccc=0. 564 * (Standard Unicode 10 normalization does not have such characters.) 565 */ 566 UBool isMostDecompYesAndZeroCC(uint16_t norm16) const { 567 return norm16<minYesNo || norm16==MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT; 568 } 569 UBool isDecompNoAlgorithmic(uint16_t norm16) const { return norm16>=limitNoNo; } 570 571 // For use with isCompYes(). 572 // Perhaps the compiler can combine the two tests for MIN_YES_YES_WITH_CC. 573 // static uint8_t getCCFromYes(uint16_t norm16) { 574 // return norm16>=MIN_YES_YES_WITH_CC ? getCCFromNormalYesOrMaybe(norm16) : 0; 575 // } 576 uint8_t getCCFromNoNo(uint16_t norm16) const { 577 const uint16_t *mapping=getMapping(norm16); 578 if(*mapping&MAPPING_HAS_CCC_LCCC_WORD) { 579 return (uint8_t)*(mapping-1); 580 } else { 581 return 0; 582 } 583 } 584 // requires that the [cpStart..cpLimit[ character passes isCompYesAndZeroCC() 585 uint8_t getTrailCCFromCompYesAndZeroCC(uint16_t norm16) const { 586 if(norm16<=minYesNo) { 587 return 0; // yesYes and Hangul LV have ccc=tccc=0 588 } else { 589 // For Hangul LVT we harmlessly fetch a firstUnit with tccc=0 here. 590 return (uint8_t)(*getMapping(norm16)>>8); // tccc from yesNo 591 } 592 } 593 uint8_t getPreviousTrailCC(const UChar *start, const UChar *p) const; 594 uint8_t getPreviousTrailCC(const uint8_t *start, const uint8_t *p) const; 595 596 // Requires algorithmic-NoNo. 597 UChar32 mapAlgorithmic(UChar32 c, uint16_t norm16) const { 598 return c+(norm16>>DELTA_SHIFT)-centerNoNoDelta; 599 } 600 UChar32 getAlgorithmicDelta(uint16_t norm16) const { 601 return (norm16>>DELTA_SHIFT)-centerNoNoDelta; 602 } 603 604 // Requires minYesNo<norm16<limitNoNo. 605 const uint16_t *getMapping(uint16_t norm16) const { return extraData+(norm16>>OFFSET_SHIFT); } 606 const uint16_t *getCompositionsListForDecompYes(uint16_t norm16) const { 607 if(norm16<JAMO_L || MIN_NORMAL_MAYBE_YES<=norm16) { 608 return NULL; 609 } else if(norm16<minMaybeYes) { 610 return getMapping(norm16); // for yesYes; if Jamo L: harmless empty list 611 } else { 612 return maybeYesCompositions+norm16-minMaybeYes; 613 } 614 } 615 const uint16_t *getCompositionsListForComposite(uint16_t norm16) const { 616 // A composite has both mapping & compositions list. 617 const uint16_t *list=getMapping(norm16); 618 return list+ // mapping pointer 619 1+ // +1 to skip the first unit with the mapping length 620 (*list&MAPPING_LENGTH_MASK); // + mapping length 621 } 622 const uint16_t *getCompositionsListForMaybe(uint16_t norm16) const { 623 // minMaybeYes<=norm16<MIN_NORMAL_MAYBE_YES 624 return maybeYesCompositions+((norm16-minMaybeYes)>>OFFSET_SHIFT); 625 } 626 /** 627 * @param c code point must have compositions 628 * @return compositions list pointer 629 */ 630 const uint16_t *getCompositionsList(uint16_t norm16) const { 631 return isDecompYes(norm16) ? 632 getCompositionsListForDecompYes(norm16) : 633 getCompositionsListForComposite(norm16); 634 } 635 636 const UChar *copyLowPrefixFromNulTerminated(const UChar *src, 637 UChar32 minNeedDataCP, 638 ReorderingBuffer *buffer, 639 UErrorCode &errorCode) const; 640 const UChar *decomposeShort(const UChar *src, const UChar *limit, 641 UBool stopAtCompBoundary, UBool onlyContiguous, 642 ReorderingBuffer &buffer, UErrorCode &errorCode) const; 643 UBool decompose(UChar32 c, uint16_t norm16, 644 ReorderingBuffer &buffer, UErrorCode &errorCode) const; 645 646 const uint8_t *decomposeShort(const uint8_t *src, const uint8_t *limit, 647 UBool stopAtCompBoundary, UBool onlyContiguous, 648 ReorderingBuffer &buffer, UErrorCode &errorCode) const; 649 650 static int32_t combine(const uint16_t *list, UChar32 trail); 651 void addComposites(const uint16_t *list, UnicodeSet &set) const; 652 void recompose(ReorderingBuffer &buffer, int32_t recomposeStartIndex, 653 UBool onlyContiguous) const; 654 655 UBool hasCompBoundaryBefore(UChar32 c, uint16_t norm16) const { 656 return c<minCompNoMaybeCP || norm16HasCompBoundaryBefore(norm16); 657 } 658 UBool norm16HasCompBoundaryBefore(uint16_t norm16) const { 659 return norm16 < minNoNoCompNoMaybeCC || isAlgorithmicNoNo(norm16); 660 } 661 UBool hasCompBoundaryBefore(const UChar *src, const UChar *limit) const; 662 UBool hasCompBoundaryBefore(const uint8_t *src, const uint8_t *limit) const; 663 UBool hasCompBoundaryAfter(const UChar *start, const UChar *p, 664 UBool onlyContiguous) const; 665 UBool hasCompBoundaryAfter(const uint8_t *start, const uint8_t *p, 666 UBool onlyContiguous) const; 667 UBool norm16HasCompBoundaryAfter(uint16_t norm16, UBool onlyContiguous) const { 668 return (norm16 & HAS_COMP_BOUNDARY_AFTER) != 0 && 669 (!onlyContiguous || isTrailCC01ForCompBoundaryAfter(norm16)); 670 } 671 /** For FCC: Given norm16 HAS_COMP_BOUNDARY_AFTER, does it have tccc<=1? */ 672 UBool isTrailCC01ForCompBoundaryAfter(uint16_t norm16) const { 673 return isInert(norm16) || (isDecompNoAlgorithmic(norm16) ? 674 (norm16 & DELTA_TCCC_MASK) <= DELTA_TCCC_1 : *getMapping(norm16) <= 0x1ff); 675 } 676 677 const UChar *findPreviousCompBoundary(const UChar *start, const UChar *p, UBool onlyContiguous) const; 678 const UChar *findNextCompBoundary(const UChar *p, const UChar *limit, UBool onlyContiguous) const; 679 680 const UChar *findPreviousFCDBoundary(const UChar *start, const UChar *p) const; 681 const UChar *findNextFCDBoundary(const UChar *p, const UChar *limit) const; 682 683 void makeCanonIterDataFromNorm16(UChar32 start, UChar32 end, const uint16_t norm16, 684 CanonIterData &newData, UErrorCode &errorCode) const; 685 686 int32_t getCanonValue(UChar32 c) const; 687 const UnicodeSet &getCanonStartSet(int32_t n) const; 688 689 // UVersionInfo dataVersion; 690 691 // BMP code point thresholds for quick check loops looking at single UTF-16 code units. 692 UChar minDecompNoCP; 693 UChar minCompNoMaybeCP; 694 UChar minLcccCP; 695 696 // Norm16 value thresholds for quick check combinations and types of extra data. 697 uint16_t minYesNo; 698 uint16_t minYesNoMappingsOnly; 699 uint16_t minNoNo; 700 uint16_t minNoNoCompBoundaryBefore; 701 uint16_t minNoNoCompNoMaybeCC; 702 uint16_t minNoNoEmpty; 703 uint16_t limitNoNo; 704 uint16_t centerNoNoDelta; 705 uint16_t minMaybeYes; 706 707 const UTrie2 *normTrie; 708 const uint16_t *maybeYesCompositions; 709 const uint16_t *extraData; // mappings and/or compositions for yesYes, yesNo & noNo characters 710 const uint8_t *smallFCD; // [0x100] one bit per 32 BMP code points, set if any FCD!=0 711 712 UInitOnce fCanonIterDataInitOnce; 713 CanonIterData *fCanonIterData; 714 }; 715 716 // bits in canonIterData 717 #define CANON_NOT_SEGMENT_STARTER 0x80000000 718 #define CANON_HAS_COMPOSITIONS 0x40000000 719 #define CANON_HAS_SET 0x200000 720 #define CANON_VALUE_MASK 0x1fffff 721 722 /** 723 * ICU-internal shortcut for quick access to standard Unicode normalization. 724 */ 725 class U_COMMON_API Normalizer2Factory { 726 public: 727 static const Normalizer2 *getFCDInstance(UErrorCode &errorCode); 728 static const Normalizer2 *getFCCInstance(UErrorCode &errorCode); 729 static const Normalizer2 *getNoopInstance(UErrorCode &errorCode); 730 731 static const Normalizer2 *getInstance(UNormalizationMode mode, UErrorCode &errorCode); 732 733 static const Normalizer2Impl *getNFCImpl(UErrorCode &errorCode); 734 static const Normalizer2Impl *getNFKCImpl(UErrorCode &errorCode); 735 static const Normalizer2Impl *getNFKC_CFImpl(UErrorCode &errorCode); 736 737 // Get the Impl instance of the Normalizer2. 738 // Must be used only when it is known that norm2 is a Normalizer2WithImpl instance. 739 static const Normalizer2Impl *getImpl(const Normalizer2 *norm2); 740 private: 741 Normalizer2Factory(); // No instantiation. 742 }; 743 744 U_NAMESPACE_END 745 746 U_CAPI int32_t U_EXPORT2 747 unorm2_swap(const UDataSwapper *ds, 748 const void *inData, int32_t length, void *outData, 749 UErrorCode *pErrorCode); 750 751 /** 752 * Get the NF*_QC property for a code point, for u_getIntPropertyValue(). 753 * @internal 754 */ 755 U_CFUNC UNormalizationCheckResult 756 unorm_getQuickCheck(UChar32 c, UNormalizationMode mode); 757 758 /** 759 * Gets the 16-bit FCD value (lead & trail CCs) for a code point, for u_getIntPropertyValue(). 760 * @internal 761 */ 762 U_CFUNC uint16_t 763 unorm_getFCD16(UChar32 c); 764 765 /** 766 * Format of Normalizer2 .nrm data files. 767 * Format version 3.0. 768 * 769 * Normalizer2 .nrm data files provide data for the Unicode Normalization algorithms. 770 * ICU ships with data files for standard Unicode Normalization Forms 771 * NFC and NFD (nfc.nrm), NFKC and NFKD (nfkc.nrm) and NFKC_Casefold (nfkc_cf.nrm). 772 * Custom (application-specific) data can be built into additional .nrm files 773 * with the gennorm2 build tool. 774 * ICU ships with one such file, uts46.nrm, for the implementation of UTS #46. 775 * 776 * Normalizer2.getInstance() causes a .nrm file to be loaded, unless it has been 777 * cached already. Internally, Normalizer2Impl.load() reads the .nrm file. 778 * 779 * A .nrm file begins with a standard ICU data file header 780 * (DataHeader, see ucmndata.h and unicode/udata.h). 781 * The UDataInfo.dataVersion field usually contains the Unicode version 782 * for which the data was generated. 783 * 784 * After the header, the file contains the following parts. 785 * Constants are defined as enum values of the Normalizer2Impl class. 786 * 787 * Many details of the data structures are described in the design doc 788 * which is at http://site.icu-project.org/design/normalization/custom 789 * 790 * int32_t indexes[indexesLength]; -- indexesLength=indexes[IX_NORM_TRIE_OFFSET]/4; 791 * 792 * The first eight indexes are byte offsets in ascending order. 793 * Each byte offset marks the start of the next part in the data file, 794 * and the end of the previous one. 795 * When two consecutive byte offsets are the same, then the corresponding part is empty. 796 * Byte offsets are offsets from after the header, 797 * that is, from the beginning of the indexes[]. 798 * Each part starts at an offset with proper alignment for its data. 799 * If necessary, the previous part may include padding bytes to achieve this alignment. 800 * 801 * minDecompNoCP=indexes[IX_MIN_DECOMP_NO_CP] is the lowest code point 802 * with a decomposition mapping, that is, with NF*D_QC=No. 803 * minCompNoMaybeCP=indexes[IX_MIN_COMP_NO_MAYBE_CP] is the lowest code point 804 * with NF*C_QC=No (has a one-way mapping) or Maybe (combines backward). 805 * minLcccCP=indexes[IX_MIN_LCCC_CP] (index 18, new in formatVersion 3) 806 * is the lowest code point with lccc!=0. 807 * 808 * The next eight indexes are thresholds of 16-bit trie values for ranges of 809 * values indicating multiple normalization properties. 810 * They are listed here in threshold order, not in the order they are stored in the indexes. 811 * minYesNo=indexes[IX_MIN_YES_NO]; 812 * minYesNoMappingsOnly=indexes[IX_MIN_YES_NO_MAPPINGS_ONLY]; 813 * minNoNo=indexes[IX_MIN_NO_NO]; 814 * minNoNoCompBoundaryBefore=indexes[IX_MIN_NO_NO_COMP_BOUNDARY_BEFORE]; 815 * minNoNoCompNoMaybeCC=indexes[IX_MIN_NO_NO_COMP_NO_MAYBE_CC]; 816 * minNoNoEmpty=indexes[IX_MIN_NO_NO_EMPTY]; 817 * limitNoNo=indexes[IX_LIMIT_NO_NO]; 818 * minMaybeYes=indexes[IX_MIN_MAYBE_YES]; 819 * See the normTrie description below and the design doc for details. 820 * 821 * UTrie2 normTrie; -- see utrie2_impl.h and utrie2.h 822 * 823 * The trie holds the main normalization data. Each code point is mapped to a 16-bit value. 824 * Rather than using independent bits in the value (which would require more than 16 bits), 825 * information is extracted primarily via range checks. 826 * Except, format version 3 uses bit 0 for hasCompBoundaryAfter(). 827 * For example, a 16-bit value norm16 in the range minYesNo<=norm16<minNoNo 828 * means that the character has NF*C_QC=Yes and NF*D_QC=No properties, 829 * which means it has a two-way (round-trip) decomposition mapping. 830 * Values in the range 2<=norm16<limitNoNo are also directly indexes into the extraData 831 * pointing to mappings, compositions lists, or both. 832 * Value norm16==INERT (0 in versions 1 & 2, 1 in version 3) 833 * means that the character is normalization-inert, that is, 834 * it does not have a mapping, does not participate in composition, has a zero 835 * canonical combining class, and forms a boundary where text before it and after it 836 * can be normalized independently. 837 * For details about how multiple properties are encoded in 16-bit values 838 * see the design doc. 839 * Note that the encoding cannot express all combinations of the properties involved; 840 * it only supports those combinations that are allowed by 841 * the Unicode Normalization algorithms. Details are in the design doc as well. 842 * The gennorm2 tool only builds .nrm files for data that conforms to the limitations. 843 * 844 * The trie has a value for each lead surrogate code unit representing the "worst case" 845 * properties of the 1024 supplementary characters whose UTF-16 form starts with 846 * the lead surrogate. If all of the 1024 supplementary characters are normalization-inert, 847 * then their lead surrogate code unit has the trie value INERT. 848 * When the lead surrogate unit's value exceeds the quick check minimum during processing, 849 * the properties for the full supplementary code point need to be looked up. 850 * 851 * uint16_t maybeYesCompositions[MIN_NORMAL_MAYBE_YES-minMaybeYes]; 852 * uint16_t extraData[]; 853 * 854 * There is only one byte offset for the end of these two arrays. 855 * The split between them is given by the constant and variable mentioned above. 856 * In version 3, the difference must be shifted right by OFFSET_SHIFT. 857 * 858 * The maybeYesCompositions array contains compositions lists for characters that 859 * combine both forward (as starters in composition pairs) 860 * and backward (as trailing characters in composition pairs). 861 * Such characters do not occur in Unicode 5.2 but are allowed by 862 * the Unicode Normalization algorithms. 863 * If there are no such characters, then minMaybeYes==MIN_NORMAL_MAYBE_YES 864 * and the maybeYesCompositions array is empty. 865 * If there are such characters, then minMaybeYes is subtracted from their norm16 values 866 * to get the index into this array. 867 * 868 * The extraData array contains compositions lists for "YesYes" characters, 869 * followed by mappings and optional compositions lists for "YesNo" characters, 870 * followed by only mappings for "NoNo" characters. 871 * (Referring to pairs of NFC/NFD quick check values.) 872 * The norm16 values of those characters are directly indexes into the extraData array. 873 * In version 3, the norm16 values must be shifted right by OFFSET_SHIFT 874 * for accessing extraData. 875 * 876 * The data structures for compositions lists and mappings are described in the design doc. 877 * 878 * uint8_t smallFCD[0x100]; -- new in format version 2 879 * 880 * This is a bit set to help speed up FCD value lookups in the absence of a full 881 * UTrie2 or other large data structure with the full FCD value mapping. 882 * 883 * Each smallFCD bit is set if any of the corresponding 32 BMP code points 884 * has a non-zero FCD value (lccc!=0 or tccc!=0). 885 * Bit 0 of smallFCD[0] is for U+0000..U+001F. Bit 7 of smallFCD[0xff] is for U+FFE0..U+FFFF. 886 * A bit for 32 lead surrogates is set if any of the 32k corresponding 887 * _supplementary_ code points has a non-zero FCD value. 888 * 889 * This bit set is most useful for the large blocks of CJK characters with FCD=0. 890 * 891 * Changes from format version 1 to format version 2 --------------------------- 892 * 893 * - Addition of data for raw (not recursively decomposed) mappings. 894 * + The MAPPING_NO_COMP_BOUNDARY_AFTER bit in the extraData is now also set when 895 * the mapping is to an empty string or when the character combines-forward. 896 * This subsumes the one actual use of the MAPPING_PLUS_COMPOSITION_LIST bit which 897 * is then repurposed for the MAPPING_HAS_RAW_MAPPING bit. 898 * + For details see the design doc. 899 * - Addition of indexes[IX_MIN_YES_NO_MAPPINGS_ONLY] and separation of the yesNo extraData into 900 * distinct ranges (combines-forward vs. not) 901 * so that a range check can be used to find out if there is a compositions list. 902 * This is fully equivalent with formatVersion 1's MAPPING_PLUS_COMPOSITION_LIST flag. 903 * It is needed for the new (in ICU 49) composePair(), not for other normalization. 904 * - Addition of the smallFCD[] bit set. 905 * 906 * Changes from format version 2 to format version 3 (ICU 60) ------------------ 907 * 908 * - norm16 bit 0 indicates hasCompBoundaryAfter(), 909 * except that for contiguous composition (FCC) the tccc must be checked as well. 910 * Data indexes and ccc values are shifted left by one (OFFSET_SHIFT). 911 * Thresholds like minNoNo are tested before shifting. 912 * 913 * - Algorithmic mapping deltas are shifted left by two more bits (total DELTA_SHIFT), 914 * to make room for two bits (three values) indicating whether the tccc is 0, 1, or greater. 915 * See DELTA_TCCC_MASK etc. 916 * This helps with fetching tccc/FCD values and FCC hasCompBoundaryAfter(). 917 * minMaybeYes is 8-aligned so that the DELTA_TCCC_MASK bits can be tested directly. 918 * 919 * - Algorithmic mappings are only used for mapping to "comp yes and ccc=0" characters, 920 * and ASCII characters are mapped algorithmically only to other ASCII characters. 921 * This helps with hasCompBoundaryBefore() and compose() fast paths. 922 * It is never necessary any more to loop for algorithmic mappings. 923 * 924 * - Addition of indexes[IX_MIN_NO_NO_COMP_BOUNDARY_BEFORE], 925 * indexes[IX_MIN_NO_NO_COMP_NO_MAYBE_CC], and indexes[IX_MIN_NO_NO_EMPTY], 926 * and separation of the noNo extraData into distinct ranges. 927 * With this, the noNo norm16 value indicates whether the mapping is 928 * compose-normalized, not normalized but hasCompBoundaryBefore(), 929 * not even that, or maps to an empty string. 930 * hasCompBoundaryBefore() can be determined solely from the norm16 value. 931 * 932 * - The norm16 value for Hangul LVT is now different from that for Hangul LV, 933 * so that hasCompBoundaryAfter() need not check for the syllable type. 934 * For Hangul LV, minYesNo continues to be used (no comp-boundary-after). 935 * For Hangul LVT, minYesNoMappingsOnly|HAS_COMP_BOUNDARY_AFTER is used. 936 * The extraData units at these indexes are set to firstUnit=2 and firstUnit=3, respectively, 937 * to simplify some code. 938 * 939 * - The extraData firstUnit bit 5 is no longer necessary 940 * (norm16 bit 0 used instead of firstUnit MAPPING_NO_COMP_BOUNDARY_AFTER), 941 * is reserved again, and always set to 0. 942 * 943 * - Addition of indexes[IX_MIN_LCCC_CP], the first code point where lccc!=0. 944 * This used to be hardcoded to U+0300, but in data like NFKC_Casefold it is lower: 945 * U+00AD Soft Hyphen maps to an empty string, 946 * which is artificially assigned "worst case" values lccc=1 and tccc=255. 947 * 948 * - A mapping to an empty string has explicit lccc=1 and tccc=255 values. 949 */ 950 951 #endif /* !UCONFIG_NO_NORMALIZATION */ 952 #endif /* __NORMALIZER2IMPL_H__ */ 953